Sebastian Raschka • 6/18/2026

GLM-5.2 and IndexShare for Long-Context Sparse Attention

This article reviews GLM-5.2, the latest open-weight model from Z.ai, highlighting its architecture built on GLM-5 and GLM-5.1. It introduces IndexShare, a cross-layer reuse trick for DeepSeek Sparse Attention (DSA) that reduces computational cost by running the top-k indexer only every four layers instead of every layer, enabling cheaper 1M-token inference. The model achieves impressive coding benchmark scores, outperforming Claude Opus 4.8. The article includes architecture details, config links, and benchmark comparisons.

0 comments

#Long Context #Sparse Attention #Glm 52